Raytrix Light Field SDK  4.0
Logo
Public Member Functions | List of all members
Rx::LFR::Net::CudaCompute Class Reference

Detailed Description

Provides methods for using a CUDA device as the computation unit of the LFR.

Inherits Rx::Net::NativeInterfaceWrapper< Rx::LFR::CCudaCompute, Rx::LFR::Net::Interfaces::ECudaCompute >.

Public Member Functions

 CudaCompute ()
 Default constructor. More...
 
 CudaCompute (Rx::LFR::CCudaCompute &xNative)
 Constructor that wraps the given native instance. Disposing does NOT destroy the native instance. More...
 
 ~CudaCompute ()
 Destructor. More...
 
void ApplyCalibration (Calibration^ xCalib, bool bUploadGrayImg)
 Applies the given calibration to the algorithms. This must be done every time the calibration changes. More...
 
void ApplyRayImage (Rx::LFR::Net::RayImage^ xRayImage)
 Applies the given ray image. More...
 
void Compute_Depth3D ()
 Calculates the depth 3D image. More...
 
void Compute_DepthColorCode (ESpace eSpace)
 Calculates the colored depth map image. More...
 
void Compute_DepthMap (ESpace eSpace)
 Calculates the depth map image. This image can be filled to produce a complete filled depth map. More...
 
void Compute_DepthRay ()
 Estimates the depth of the current raw image and creates the ray depth image. More...
 
void Compute_Grid ()
 Calculates the grid image. This is the raw or the gray image with an overlay that shows the current grid calibration. More...
 
void Compute_GridDataImage ()
 Pre-Process a ray image with a minimal set of preprocessing parameters. More...
 
void Compute_Histogram (EImage eImgID, RX_OUT array< unsigned >^%xHistogram, RX_OUT unsigned% uChannels)
 Calculates the histogram of the image with the given image ID and stores the histogram data in the given array. The number of calculated channels is returned in uChannels. More...
 
void Compute_ImageMinMax (EImage eImgID, unsigned uChannel, RX_OUT float% fMin, RX_OUT float% fMax)
 Calculates the minimum and the maximum value in the given image. More...
 
void Compute_Multiview ()
 Calculates the multi view image. More...
 
void Compute_PreProcess ()
 Pre-Process a ray image. This can apply denoise, brightness and sharpness adjustments. More...
 
void Compute_RefocusBasic ()
 Focus image to a plane perpendicular to the optical axis, i.e. parallel to the image plane. More...
 
void Compute_TotalFocus (ESpace eSpace)
 Focus on a depth surface. More...
 
Rx::Net::ImageFormatCreateFocusImageFormat (Rx::Net::ImageFormat^ xRawFormat, bool bForceLandscape)
 Creates the image format of the refocus and the total focus image. This considers the Focus_ImageDivisor parameter. More...
 
CudaDeviceGetCudaDevice ()
 Gets the CUDA device that has been set by calling SetCudaDevice. Throws an exception if not set yet. More...
 
virtual System::ObjectGetInterface (Interfaces::ECudaCompute eInterface) override
 Gets the interface of the CUDA compute class determined by the Interfaces::CudaCompute enum. More...
 
IParameters< Rx::LFR::Net::Params::ECudaCompute >^ GetParams ()
 Gets the parameters access interface. More...
 
void ReinitializeDepthDB ()
 Reinitializes the depth database. More...
 
void RemoveComputationFilterMask ()
 Removes the computation filter mask set by SetComputationFilterMask. More...
 
void ResizeImage (Rx::Net::Image^ xTrgImg, Rx::Net::Image^ xSrcImg, int iDivisor)
 Resize the given source image and store the result in the given target image using the given divisor. More...
 
void SetComputationFilterMask (Rx::Net::Image^ xImage)
 Sets the computation filter mask image. More...
 
void SetComputationFilterMask (System::String^ sFilename)
 Sets the computation filter mask image. More...
 
void SetCudaDevice (CudaDevice^ xCudaDevice)
 Sets the CUDA device that allocates the memory required for the operations and that runs the algorithms. More...
 
void UploadRawImage (Rx::Net::Image^ xSrcImg)
 Uploads the given image as the new raw image of all computations. More...
 

Constructor & Destructor Documentation

Rx::LFR::Net::CudaCompute::CudaCompute ( )
inline

Default constructor.

Rx::LFR::Net::CudaCompute::CudaCompute ( Rx::LFR::CCudaCompute xNative)
inline

Constructor that wraps the given native instance. Disposing does NOT destroy the native instance.

Parameters
xNative[in] The native instance to wrap by this class.
Rx::LFR::Net::CudaCompute::~CudaCompute ( )
inline

Destructor.

Member Function Documentation

void Rx::LFR::Net::CudaCompute::ApplyCalibration ( Calibration xCalib,
bool  bUploadGrayImg 
)
inline

Applies the given calibration to the algorithms. This must be done every time the calibration changes.

Parameters
xCalib[in] The calibration.
bUploadGrayImgTrue to upload the gray image of the calibration to CUDA. False to use the current gray image even if it is invalid.
void Rx::LFR::Net::CudaCompute::ApplyRayImage ( Rx::LFR::Net::RayImage xRayImage)
inline

Applies the given ray image.

This uploads the raw and the gray image to the CUDA device and applies the calibration and the parameters from the meta data to the CUDA algorithms.

Parameters
xRayImageThe ray image.
void Rx::LFR::Net::CudaCompute::Compute_Depth3D ( )
inline

Calculates the depth 3D image.

This creates a RGBA float image which maps each pixel in the view camera to a 3D position in object space. The R channel is the X coordinate, the G channel is the Y coordinate and the B channel is the Z coordinate. The Z coordinate is relative to the global coordinate system which typically coincides with the light field camera sensor. All coordinate values are given in millimeters. The A channel is the Z coordinate in virtual depths.

Result image:

Required images:

Influencing parameters:

  • none.
void Rx::LFR::Net::CudaCompute::Compute_DepthColorCode ( ESpace  eSpace)
inline

Calculates the colored depth map image.

Result image:

Required images:

Influencing parameters:

  • ECCPar::DepthMap_Color_Mode
  • ECCPar::DepthMap_Color_Max
  • ECCPar::DepthMap_Color_Min
  • ECCPar::DepthMap_Color_Map.
Parameters
eSpaceThe space. This is LF::ESpace::View_Virtual or LF::ESpace::View_Object_Pinhole.
void Rx::LFR::Net::CudaCompute::Compute_DepthMap ( ESpace  eSpace)
inline

Calculates the depth map image. This image can be filled to produce a complete filled depth map.

Result image:

Required images:

Influencing parameters:

  • ECCPar::Depth_ConsistencyCheck
  • ECCPar::Depth_Fill_Enabled
  • ECCPar::Depth_Fill_IterCnt
  • ECCPar::Depth_Fill_IterSize
  • ECCPar::Depth_Fill_Complete
  • ECCPar::Depth_Fill_ShrinkRangePX
  • ECCPar::Depth_Fill_Filter_Enabled
  • ECCPar::Depth_Fill_Bilateral_Enabled
  • ECCPar::Depth_Fill_Bilateral_Edge
  • ECCPar::Depth_Fill_Bilateral_Range
  • ECCPar::Depth_Fill_Bilateral_FilterRadius.
Parameters
eSpaceThe space. This is LF::ESpace::View_Virtual or LF::ESpace::View_Object_Pinhole.
void Rx::LFR::Net::CudaCompute::Compute_DepthRay ( )
inline

Estimates the depth of the current raw image and creates the ray depth image.

Result image:

Required images:

Influencing parameters:

  • ECCPar::PreProc_Sharp1_Enable
  • ECCPar::PreProc_Sharp1_OnlyCalc
  • ECCPar::PreProc_Sharp1_BlurStdDev
  • ECCPar::PreProc_Sharp1_Factor
  • ECCPar::Depth_SparseLensSet
  • ECCPar::Depth_NearResolutionLevel
  • ECCPar::Depth_PatchDiameter
  • ECCPar::Depth_PatchStride
  • ECCPar::Depth_MinVirtualDepth
  • ECCPar::Depth_MaxVirtualDepth
  • ECCPar::Depth_MinCorrelation
  • ECCPar::Depth_MinStdDeviation
  • ECCPar::Depth_LensFilterMask.
void Rx::LFR::Net::CudaCompute::Compute_Grid ( )
inline

Calculates the grid image. This is the raw or the gray image with an overlay that shows the current grid calibration.

Result image:

Required images:

Influencing parameters:

  • ECCPar::Grid_DrawOnGrayImage
  • ECCPar::Grid_DrawMlaGrid.
void Rx::LFR::Net::CudaCompute::Compute_GridDataImage ( )
inline

Pre-Process a ray image with a minimal set of preprocessing parameters.

Result image:

Required images:

Influencing parameters:

  • ECCPar::PreProc_Demosaic_R
  • ECCPar::PreProc_Demosaic_G
  • ECCPar::PreProc_Demosaic_B.
void Rx::LFR::Net::CudaCompute::Compute_Histogram ( EImage  eImgID,
RX_OUT array< unsigned >^%  xHistogram,
RX_OUT unsigned%  uChannels 
)
inline

Calculates the histogram of the image with the given image ID and stores the histogram data in the given array. The number of calculated channels is returned in uChannels.

Parameters
eImgIDThe image ID.
xHistogram[out] The histogram.
uChannels[out] The number of channels.
void Rx::LFR::Net::CudaCompute::Compute_ImageMinMax ( EImage  eImgID,
unsigned  uChannel,
RX_OUT float%  fMin,
RX_OUT float%  fMax 
)
inline

Calculates the minimum and the maximum value in the given image.

Parameters
eImgIDThe image ID. The image must be of type float.
uChannelThe channel of each pixel that is used to get the values.
fMin[out] The minimum.
fMax[out] The maximum.
void Rx::LFR::Net::CudaCompute::Compute_Multiview ( )
inline

Calculates the multi view image.

Result image:

Required images:

Influencing parameters:

  • ECCPar::Multiview_Type
  • ECCPar::Multiview_EyeSeparation
  • ECCPar::Depth_ConsistencyCheck
  • ECCPar::Depth_Fill_Enabled
  • ECCPar::Depth_Fill_IterCnt
  • ECCPar::Depth_Fill_IterSize
  • ECCPar::Depth_Fill_Complete
  • ECCPar::Depth_Fill_ShrinkRangePX
  • ECCPar::Depth_Fill_Filter_Enabled
  • ECCPar::Depth_Fill_Bilateral_Enabled
  • ECCPar::Depth_Fill_Bilateral_Edge
  • ECCPar::Depth_Fill_Bilateral_Range
  • ECCPar::Depth_Fill_Bilateral_FilterRadius.
void Rx::LFR::Net::CudaCompute::Compute_PreProcess ( )
inline

Pre-Process a ray image. This can apply denoise, brightness and sharpness adjustments.

Result image:

Required images:

Influencing parameters:

  • ECCPar::PreProc_Demosaic_Enable
  • ECCPar::PreProc_Demosaic_R
  • ECCPar::PreProc_Demosaic_G
  • ECCPar::PreProc_Demosaic_B
  • ECCPar::PreProc_DenoiseNLM_Enable
  • ECCPar::PreProc_DenoiseNLM_FilterDia
  • ECCPar::PreProc_DenoiseNLM_NoiseLevel
  • ECCPar::PreProc_DenoiseNLM_BlendFactor
  • ECCPar::PreProc_Std_Enable
  • ECCPar::PreProc_Std_Gradation
  • ECCPar::PreProc_Sharp1_Enable
  • ECCPar::PreProc_Sharp1_OnlyCalc
  • ECCPar::PreProc_Sharp1_BlurStdDev
  • ECCPar::PreProc_Sharp1_Factor.
void Rx::LFR::Net::CudaCompute::Compute_RefocusBasic ( )
inline

Focus image to a plane perpendicular to the optical axis, i.e. parallel to the image plane.

This focuses to a plane perpendicular to the optical axis. You can change the relative focal plane. A relative focus value of zero refocuses to the plane furthest away from the viewer and a value of 1 would focus to a plane that passes through the object side focal point of the recording lens.

Result image:

Required images:

Influencing parameters:

  • ECCPar::Focus_RelativeFocusPlane
  • ECCPar::Depth_MinVirtualDepth
  • ECCPar::Depth_MaxVirtualDepth.
void Rx::LFR::Net::CudaCompute::Compute_TotalFocus ( ESpace  eSpace)
inline

Focus on a depth surface.

In principle, each pixel of the result image can be focused individually. This function makes this property available by focusing the ray image to the depth surface given in the internal depth image related to the given space.

Result image:

Required images:

Influencing parameters:

  • ECCPar::Focus_RelativeFocusPlane
  • ECCPar::Focus_DepthBlendingScale
  • ECCPar::Depth_MinVirtualDepth
  • ECCPar::Depth_MaxVirtualDepth.
Parameters
eSpaceThe space. This is LF::ESpace::View_Virtual or LF::ESpace::View_Object_Pinhole.
Rx::Net::ImageFormat ^ Rx::LFR::Net::CudaCompute::CreateFocusImageFormat ( Rx::Net::ImageFormat xRawFormat,
bool  bForceLandscape 
)
inline

Creates the image format of the refocus and the total focus image. This considers the Focus_ImageDivisor parameter.

Parameters
xRawFormatThe raw image format.
bForceLandscapeTrue to force landscape. Otherwise the calibration is used.
Returns
The image format.
CudaDevice ^ Rx::LFR::Net::CudaCompute::GetCudaDevice ( )
inline

Gets the CUDA device that has been set by calling SetCudaDevice. Throws an exception if not set yet.

Returns
The CUDA device.
virtual System::Object ^ Rx::LFR::Net::CudaCompute::GetInterface ( Interfaces::ECudaCompute  eInterface)
inlineoverridevirtual

Gets the interface of the CUDA compute class determined by the Interfaces::CudaCompute enum.

Parameters
eInterfaceThe interface enum.
Returns
Null if no interface is implemented yet, else the interface object.

Reimplemented from Rx::Net::NativeInterfaceWrapper< Rx::LFR::CCudaCompute, Rx::LFR::Net::Interfaces::ECudaCompute >.

IParameters<Rx::LFR::Net::Params::ECudaCompute> ^ Rx::LFR::Net::CudaCompute::GetParams ( )
inline

Gets the parameters access interface.

Returns
The parameters access interface.
void Rx::LFR::Net::CudaCompute::ReinitializeDepthDB ( )
inline

Reinitializes the depth database.

This can be useful when calling Compute_DepthMap after changing the ray depth image manually.

void Rx::LFR::Net::CudaCompute::RemoveComputationFilterMask ( )
inline

Removes the computation filter mask set by SetComputationFilterMask.

void Rx::LFR::Net::CudaCompute::ResizeImage ( Rx::Net::Image xTrgImg,
Rx::Net::Image xSrcImg,
int  iDivisor 
)
inline

Resize the given source image and store the result in the given target image using the given divisor.

Parameters
xTrgImg[in] The target image. This must not be a moniker. It's created inside.
xSrcImg[in] The source image.
iDivisorThe divisor. A value of 2 means that the target image is half the source size.
void Rx::LFR::Net::CudaCompute::SetComputationFilterMask ( Rx::Net::Image xImage)
inline

Sets the computation filter mask image.

This image is used to reduce the number of lenses used for depth estimation.

It must have the same dimension as the raw image and a pixel type of Rx::InteropNet::Runtime28::EPixelType::Lum and a data type of Rx::InteropNet::Runtime28::EDataType::UByte. Otherwise the image is converted.

Each lens of the raw image has a lens center that can be rounded to a pixel. If this pixel has the value 255 in the filter mask image, the corresponding lens is used. If this pixel has another value, the lens is not used.

Parameters
xImage[in] The computation filter mask image.
void Rx::LFR::Net::CudaCompute::SetComputationFilterMask ( System::String^  sFilename)
inline

Sets the computation filter mask image.

This image is used to reduce the number of lenses used for depth estimation.

It must have the same dimension as the raw image and a pixel type of Rx::InteropNet::Runtime28::EPixelType::Lum and a data type of Rx::InteropNet::Runtime28::EDataType::UByte. Otherwise the image is converted.

Each lens of the raw image has a lens center that can be rounded to a pixel. If this pixel has the value 255 in the filter mask image, the corresponding lens is used. If this pixel has another value, the lens is not used.

Parameters
sFilename[in] The complete image file path.
void Rx::LFR::Net::CudaCompute::SetCudaDevice ( CudaDevice xCudaDevice)
inline

Sets the CUDA device that allocates the memory required for the operations and that runs the algorithms.

Parameters
xCudaDevice[in] The CUDA device.
void Rx::LFR::Net::CudaCompute::UploadRawImage ( Rx::Net::Image xSrcImg)
inline

Uploads the given image as the new raw image of all computations.

Parameters
xSrcImg[in] The source image.